Kevin DeLuca is an Assistant Professor of Political Science at Yale University. He received his PhD in Political Economy from Harvard in 2023 and is an IQSS alumnus. Here, he describes political endorsement data that he collected, with the help of IQSS graduate research funding, for use in his research and PhD dissertation.
Political scientists have long recognized the importance of “candidate quality” in explaining electoral politics. Candidate quality refers to attributes of political candidates that make them better at governing—such as competency, experience, or not being corrupt. Candidate quality matters because higher quality politicians improve the welfare of voters, regardless of their partisan preferences. In theory, voters would prefer to have high quality representatives (all else equal) because higher quality leaders make government more efficient and valuable to citizens.
Empirically, however, it is hard to quantify the effect that candidate quality has on political outcomes. That’s because candidate quality itself is difficult to measure. To overcome this limitation, I use local newspaper endorsements of political candidates to estimate a new measure of candidate quality. The simple idea is that local newspapers are more likely to endorse candidates who are high quality, after taking into account the newspaper’s partisan bias. In my research, I use this novel measure of candidate quality to assess the impact that quality has on elections and governance.[1]
In the United States, it is common for editors of local newspapers to make endorsements of candidates leading up to election day. Newspaper editors cover elections extensively, know much more than a typical voter about the issues and the candidates running, and have experience assessing candidate performance. Often, they interview candidates, especially for more local offices. They provide endorsements as a service to their readers, who do not have the time or resources to learn about the election in the same way that newspaper reporters do, and they often publish an editorial justifying their endorsements to their audience. These local newspaper endorsements can be thought of as “expert opinions” of the candidates that are, in part, based on each candidates’ quality.
However, newspapers may be biased—they may prefer one political party over the other. Endorsements also reflect the partisan bias of the newspapers making the endorsements, not just candidate quality. Therefore, it is important to take into account the partisan preferences of newspapers when using their endorsements as signals of candidate quality. On the other hand, the bias of the newspapers is useful when measuring candidate quality because it helps reveal information about the magnitude of any differences in candidate quality. Newspapers that are very biased will only deviate from their typical pattern of endorsements when the quality differences between candidates are large. For example, if a Democratic candidate earns many newspaper endorsements, including endorsements from newspapers that typically prefer Republicans, that is a strong signal that the Democrat is a high-quality candidate (or, that the Republican candidate is low-quality).
With a large enough set of newspaper endorsements from a single newspaper, one could estimate the partisan bias of that newspaper; and with a large enough set of newspaper endorsements in a particular election—coming from many different newspapers with many different biases—one could estimate not only which of two candidates was higher quality, but also the magnitude of that difference in quality.
To estimate candidate quality empirically, I use a large dataset of over 22,000 local newspaper endorsements of political candidates. This dataset was manually collected throughout my time in graduate school with support from IQSS, with contributions from Jim Snyder, Tyler Simko, David Beavers, Matthew Kind, Patricia Hughes, along with a handful of other research assistants over many years. The endorsements come from over 400 different newspapers, spanning the years 1950-2022, and include endorsements in thousands of elections across all office types, from president down to local city government positions.
Presented below in Figure 1 is a typical example of a newspaper editorial endorsement recorded in the dataset. In 1978, the Wilmington Evening Journal endorsed Joe Biden for his 2nd term in the U.S. Senate. They provide the reasoning for their endorsement in the form of an editorial article, published the Friday before election day.
In the text of the article, they say how Biden has a “grasp of the issues” and “knows what he is talking about” and they point to his record of service as a reason why he should be reelected. This is typical language used in article endorsements of candidates, and suggests they are focusing on quality-related traits of candidates when making their endorsements. Additionally, in my dissertation I show that quality-related factors such as incumbency status, prior experience, years of experience, and (not) being in a scandal all predict being endorsed by local newspapers. Both qualitative and quantitative evidence, therefore, supports the idea that newspaper endorsements are indicators of candidate quality.
I use the full dataset of newspaper endorsements to simultaneously produce 1) a dynamic measure of the partisan bias of each newspaper, and 2) an estimate of the quality differences between candidates in thousands of elections across the United States. I use a high-dimensional fixed effects linear probability model, where the outcome is an indicator variable for a Democratic endorsement, and the explanatory variables are newspaper fixed effects (with a linear time trend for each newspaper) and election-specific fixed effects. The newspaper fixed effects capture how likely it is that a particular newspaper endorses a Democrat, whereas the election fixed effects capture how likely it is that the Democratic candidate in that particular election is endorsed across all papers making endorsements. In other words, the newspaper fixed effects measure the newspaper’s partisan bias, and the election fixed effects are a measure of the differences in candidate quality in that election.
Allowing each newspaper’s bias to change over time is important because the political landscape and economic incentives faced by newspapers changed significantly between 1950-2022. Figure 2 below shows the overall distribution of the partisan slant of newspapers in the 1950s and in the 2010s, with higher values of slant indicating a pro-Democratic bias in endorsement behavior. Consistent with previous research, I find that local news is strongly Republican leaning in the 1950s. In the most recent decade, however, I find that local news has generally become more Democratic leaning, perhaps in response to the anti-establishment and anti-media turn within the Republican party (exemplified by the Tea Party movement and the subsequent rise of “Trumpism”).
The full endorsement dataset includes endorsements for 6,502 unique elections; for each of these elections, I estimate the difference in candidate quality, as indicated by the likelihood of the Democratic candidate being endorsed conditional on each newspaper’s bias. I merge these differences in candidate quality—what I refer to in the paper as “candidate quality differentials”—to election results in order to assess the impact of candidate quality on elections.
The left panel of Figure 3 below shows a binned scatterplot of the relationship between candidate quality differentials and Democratic two-party vote shares in an election. With no controls, I estimate that a one standard deviation increase in candidate quality differentials is associated with an 8.6 percentage point increase in a candidate’s vote shares; when controlling for expected vote shares, incumbency, and partisan tides, the estimated effect is 3.8 percentage points. The right panel of Figure 3 demonstrates that not only do higher quality candidates win more votes, they are also more likely to win their elections (27.9 percent more likely without controls, 14.1 percent more likely with controls). This figure and the accompanying estimates reveal a strong relation between candidate quality and electoral performance.
But how do we know that electing higher quality candidates actually leads to better governing? With the help of the new endorsement-based quality differential estimated here, we can try to answer that question. To do so, I look at the relationship between quality differentials and two important measures of performance that have been widely used in political science: 1) legislative effectiveness scores and 2) net approval ratings. Each can be viewed as a somewhat objective measure of governing performance, and if electing higher quality candidates leads to better governing, we’d expect that legislator effectiveness and approval ratings would both be higher when a high-quality candidate wins.
Table 1 below reveals that indeed this is the case: candidates who are higher quality are both more effective legislators once in office and have higher net approval ratings relative to low quality candidates. This important result reveals that not only does the endorsement-based quality measure reflect traits that voters value at the ballot box, but also reflects attributes of candidates that lead to objectively better governance.
In the dissertation, I also show that the effects of candidate quality on election results and governing performance holds even when adjusting for legislator ideological extremity (when such measures are available, both using campaign finance (CF-)scores or DW-Nominate scores). Additionally, another IQSS alumnus, Shiro Kuriwaki, has shown that the endorsement-based quality differential also predicts down-ballot split ticket voting [2], further validating the use of the endorsement-based quality differential and demonstrating that voters do know and care about candidate quality, even in today’s polarized environment.
The endorsement-based measure of candidate quality differentials has a strong potential to further enhance our understanding of how the quality of politicians affects elections, public policy, and the overall quality of political representation in the United States. In other papers, I use the endorsement-based quality differentials to better understand the relationship between incumbency and candidate quality [3], whether the importance of candidate quality has decreased over time with the rise of political polarization, and whether newspaper endorsements per se affect voter decisions [4]. In future work, I plan to use the measures of newspaper partisan bias to study how media bias affects voter behavior, and the endorsement dataset itself can be used to analyze how newspaper behavior has changed over time along with the changing media landscape. Overall, these findings offer important insights and tools for scholars examining the dynamics of elections and representation in the United States.
